Dropout Concrete Autoencoder for Band Selection on HSI Scenes

Lei Xu; Mete Ahishali; Moncef Gabbouj

arxiv: 2401.16522 · v1 · submitted 2024-01-29 · 💻 cs.CV

Dropout Concrete Autoencoder for Band Selection on HSI Scenes

Lei Xu , Mete Ahishali , Moncef Gabbouj This is my paper

Pith reviewed 2026-05-24 03:48 UTC · model grok-4.3

classification 💻 cs.CV

keywords band selectionhyperspectral imagesconcrete autoencoderdropoutend-to-end learningfeature rankingspectral redundancy

0 comments

The pith

A dropout concrete autoencoder enables direct end-to-end selection of informative bands in hyperspectral images.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a neural network for choosing useful spectral bands from hyperspectral images that trains in a single end-to-end pass. Prior deep learning approaches could not directly handle the discrete choice of which bands to keep and therefore needed separate post-processing steps or indirect optimization. By merging concrete autoencoder parameterization with a dropout feature ranking strategy, the method optimizes the band subset directly for any chosen size while removing redundancies. Tests across four hyperspectral scenes indicate that the resulting selections deliver higher performance than competing methods.

Core claim

The proposed dropout concrete autoencoder is trained directly given the required band subset, eliminating the need for further post-processing, and achieves substantial and effective performance levels outperforming the competing methods on four HSI scenes.

What carries the argument

The dropout concrete autoencoder, formed by combining concrete autoencoder parameterization with dropout feature ranking to support direct optimization of discrete band choices.

If this is right

The network trains directly for any required band subset size.
No post-processing strategies are needed after training completes.
The selections outperform those from competing methods on the four tested HSI scenes.
Spectral correlation and redundancies are reduced through the single-pass optimization.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same hybrid parameterization could be tested on other discrete selection tasks that currently require separate ranking stages.
Removing post-processing steps may shorten overall processing time in hyperspectral analysis pipelines.
Evaluating the approach on additional HSI datasets beyond the four scenes would show how far the direct optimization generalizes.

Load-bearing premise

The concrete autoencoder combined with dropout feature ranking permits direct end-to-end optimization of the discrete band selection variables.

What would settle it

Training the network on the four HSI scenes and verifying whether band selection occurs without any post-processing while still exceeding baseline performance; failure on either point would refute the direct end-to-end claim.

read the original abstract

Deep learning-based informative band selection methods on hyperspectral images (HSI) recently have gained intense attention to eliminate spectral correlation and redundancies. However, the existing deep learning-based methods either need additional post-processing strategies to select the descriptive bands or optimize the model indirectly, due to the parameterization inability of discrete variables for the selection procedure. To overcome these limitations, this work proposes a novel end-to-end network for informative band selection. The proposed network is inspired by the advances in concrete autoencoder (CAE) and dropout feature ranking strategy. Different from the traditional deep learning-based methods, the proposed network is trained directly given the required band subset eliminating the need for further post-processing. Experimental results on four HSI scenes show that the proposed dropout CAE achieves substantial and effective performance levels outperforming the competing methods.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This is an incremental combination of concrete autoencoders and dropout ranking aimed at direct end-to-end band selection for hyperspectral images, but the abstract supplies zero numbers or details so the performance claims cannot be checked.

read the letter

The paper takes the concrete autoencoder and pairs it with a dropout feature ranking step to train a network that outputs the selected bands directly. The abstract says this removes the need for post-processing or indirect optimization that earlier deep learning band selection methods required. That is the main technical move: one network, trained once, that produces the subset without extra stages afterward. On four HSI scenes it claims to beat the competing methods at a substantial level. The idea is straightforward and targets a real workflow annoyance in the subfield. If the training actually converges to good selections without the usual extra steps, the approach could be practically convenient for remote sensing pipelines that already use autoencoders. The limitation is obvious from what is provided. No metrics appear, no baselines are named, no architecture or loss details are given, and no tables or statistical tests are shown. The claim of outperformance therefore sits unsupported. Because only the abstract is available, it is impossible to tell whether the combination introduces new instabilities, whether the selected bands are actually informative, or whether the gains are consistent across scenes. This kind of work would interest people already doing band selection on hyperspectral data who follow the CAE literature. A reader who needs a new end-to-end baseline might want to see the full experiments once they exist. I would not cite the paper in its present state because there is nothing concrete to reference or build on. For peer review the manuscript is too thin; an editor should ask for the methods, results, and code before sending it to referees.

Referee Report

1 major / 1 minor

Summary. The manuscript proposes a novel end-to-end network called the dropout Concrete Autoencoder (CAE) for informative band selection in hyperspectral images (HSI). Inspired by advances in CAE and dropout feature ranking, the method is trained directly on the required band subset to eliminate the need for post-processing or indirect optimization of discrete variables. The abstract claims that experiments on four HSI scenes demonstrate substantial and effective performance levels, outperforming competing methods.

Significance. If the outperformance claims hold with proper validation, the approach would address a recognized limitation in deep learning-based band selection by enabling direct end-to-end optimization of discrete selections. This could be relevant to the HSI community. However, the manuscript consists solely of the abstract and supplies no metrics, baselines, statistical tests, architecture details, or experimental results, rendering any assessment of significance impossible at present.

major comments (1)

The abstract asserts outperformance on four HSI scenes but supplies no metrics, baselines, statistical tests, or experimental details, making it impossible to assess whether the data actually support the central claim of superiority. This is load-bearing for the paper's main contribution.

minor comments (1)

The provided manuscript contains only the abstract; no methods section, equations, tables, figures, or training details are available for technical review.

Simulated Author's Rebuttal

1 responses · 1 unresolved

We thank the referee for their review. The single major comment is addressed below. We agree that the provided manuscript text is limited to the abstract and lacks supporting experimental information.

read point-by-point responses

Referee: The abstract asserts outperformance on four HSI scenes but supplies no metrics, baselines, statistical tests, or experimental details, making it impossible to assess whether the data actually support the central claim of superiority. This is load-bearing for the paper's main contribution.

Authors: We agree with this assessment. The text supplied consists solely of the abstract, which states the outperformance claim without any accompanying metrics, baselines, architecture details, or results. Without these elements, it is not possible to evaluate whether the data support the claims of superiority over competing methods. In a full manuscript, dedicated experimental sections would be required to substantiate the assertions. revision: yes

standing simulated objections not resolved

Whether the experimental data actually support the outperformance claims on four HSI scenes, as no metrics, baselines, or results are present in the provided manuscript.

Circularity Check

0 steps flagged

No significant circularity; abstract contains no derivations or equations

full rationale

The provided text is limited to the abstract, which describes a method inspired by prior CAE and dropout ideas but presents no equations, derivations, fitted parameters, or self-citations. No load-bearing steps exist that could reduce to inputs by construction. The claim of end-to-end optimization is stated at a high level without any mathematical reduction that could be inspected for circularity. This is the normal case of a self-contained description against external benchmarks when no internal chain is available to analyze.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the central claim rests on the unstated premise that the CAE-plus-dropout construction solves the discrete-variable parameterization problem.

pith-pipeline@v0.9.0 · 5637 in / 991 out tokens · 20743 ms · 2026-05-24T03:48:09.900934+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages · 2 internal anchors

[1]

Dropout Concrete Autoencoder for Band Selection on HSI Scenes

INTRODUCTION Hyperspectral images (HSI) captured by hyperspectral remote sensing imaging spectrometers [1] cover a wide and continu- ous range of the electromagnetic spectrum beyond the visible wavelengths with multiple spectral bands. Due to this charac- teristic, hyperspectral images contain enormous information utilizing its various applications, such ...

work page internal anchor Pith review Pith/arXiv arXiv 2024
[2]

Next, the pseudo-code of the proposed method is provided at the end of this section

PROPOSED METHOD In this section, we first present the principle of concrete dis- tribution, dropout feature ranking, and the proposed Dropout CAE in detail. Next, the pseudo-code of the proposed method is provided at the end of this section. 2.1. Concrete Distribution The concrete random variables are defined as a continuous re- laxation of discrete rando...

work page
[3]

EXPERIMENTAL EV ALUA TION In this section, we first describe the datasets used in the ex- periments, and we provide implementation details which are followed by comparative evaluations. 3.1. Datasets In this work, we have evaluated the proposed network using four HSI scenes: Indian Pines [24], PaviaU [25], Salinas [26], and KSC [27]. The original HSI scen...

work page
[4]

with three Q values (polynomial approximation order), SpaBS [30], EGCSR [31], ISSC [9], and BS-Net-FC [12]. Dropout CAE (T1) Dropout CAE (T2) Dropout CAE (T3) BS_Net_FC EGCSR ISSC SpaBS SRL_SOR Q1 SRL_SOR Q2 SRL_SOR Q3 0 7 15 30 45 60 75 90 105 120 135 150 165 180 195205 Spectral Band 1 2 3 4 5Value of Entropy Fig. 3: The selected subset with 20 bands and...

work page
[5]

We first utilize the variational dropout strategy to exploit the importance of each frequency band for HSI scenes

CONCLUSION In this work, we propose a novel method named Dropout CAE to re-parameterize the discrete random variables for HSI band selection. We first utilize the variational dropout strategy to exploit the importance of each frequency band for HSI scenes. To bridge the gap between the discrete band information and the re-parameterization of the discrete ...

work page
[6]

Ngadi and Li Liu, Hyperspectral Image Pro- cessing Techniques, Number December 2010

Michael O. Ngadi and Li Liu, Hyperspectral Image Pro- cessing Techniques, Number December 2010. Elsevier Inc., first edit edition, 2010

work page 2010
[7]

Hy- perspectral image data mining for band selection in agri- cultural applications,

S. G. Bajwa, P. Bajcsy, P. Groves, and L. F. Tian, “Hy- perspectral image data mining for band selection in agri- cultural applications,” Transactions of the American So- ciety of Agricultural Engineers , vol. 47, no. 3, pp. 895– 907, 2004

work page 2004
[8]

Estimation of Mineral Abundance from Hyperspectral Data Using a New Supervised Neighbor-Band Ratio Un- mixing Approach,

Kevin Siebels, Kalifa Go ¨ıta, and Micka ¨el Germain, “Estimation of Mineral Abundance from Hyperspectral Data Using a New Supervised Neighbor-Band Ratio Un- mixing Approach,” IEEE Transactions on Geoscience and Remote Sensing , vol. 58, no. 10, pp. 6754–6766, 2020

work page 2020
[9]

Advances in hyperspectral image classification: Earth monitoring with statistical learning methods,

Gustavo Camps-Valls, Devis Tuia, Lorenzo Bruzzone, and Jon Atli Benediktsson, “Advances in hyperspectral image classification: Earth monitoring with statistical learning methods,” IEEE Signal Processing Magazine , vol. 31, no. 1, pp. 45–54, 2014

work page 2014
[10]

Convolutional Neural Network Based on Bandwise- Independent Convolution and Hard Thresholding for Hyperspectral Band Selection,

Jie Feng, Jiantong Chen, Qigong Sun, Ronghua Shang, Xianghai Cao, Xiangrong Zhang, and Licheng Jiao, “Convolutional Neural Network Based on Bandwise- Independent Convolution and Hard Thresholding for Hyperspectral Band Selection,” IEEE Transactions on Cybernetics, vol. 51, no. 9, pp. 4414–4428, 2021

work page 2021
[11]

Hyperspectral Band Selection for Mul- tispectral Image Classification with Convolutional Net- works,

Giorgio Morales, John Sheppard, Riley Logan, and Joseph Shaw, “Hyperspectral Band Selection for Mul- tispectral Image Classification with Convolutional Net- works,” Proceedings of the International Joint Confer- ence on Neural Networks , vol. 2021-July, 2021

work page 2021
[12]

A joint band prioritization and band- decorrelation approach to band selection for hyperspec- tral image classification,

Chein I. Chang, “A joint band prioritization and band- decorrelation approach to band selection for hyperspec- tral image classification,” IEEE Transactions on Geo- science and Remote Sensing , vol. 37, no. 6, pp. 2631– 2641, 1999

work page 1999
[13]

Band Selection for Dimension Reduction in Hyper Spectral Image Using Integrated Information- Gain and Principal Components Analysis Technique,

Kitti Koonsanit, Chuleerat Jaruskulchai, and Apisit Eiumnoh, “Band Selection for Dimension Reduction in Hyper Spectral Image Using Integrated Information- Gain and Principal Components Analysis Technique,” International Journal of Machine Learning and Com- puting, vol. 2, no. 3, pp. 248–251, 2012

work page 2012
[14]

Band Selection Using Improved Sparse Subspace Clustering for Hyperspectral Imagery Classification,

Weiwei Sun, Liangpei Zhang, Bo Du, Weiyue Li, and Yenming Mark Lai, “Band Selection Using Improved Sparse Subspace Clustering for Hyperspectral Imagery Classification,” IEEE Journal of Selected Topics in Ap- plied Earth Observations and Remote Sensing , vol. 8, no. 6, pp. 2784–2797, 2015

work page 2015
[15]

Optimal Clustering Framework for Hyperspectral Band Selec- tion,

Qi Wang, Fahong Zhang, and Xuelong Li, “Optimal Clustering Framework for Hyperspectral Band Selec- tion,” IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 10, pp. 5910–5922, 2018

work page 2018
[16]

A Fast Neigh- borhood Grouping Method for Hyperspectral Band Se- lection,

Qi Wang, Qiang Li, and Xuelong Li, “A Fast Neigh- borhood Grouping Method for Hyperspectral Band Se- lection,” IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 6, pp. 5028–5039, 2021

work page 2021
[17]

BS-Nets: An End-to-End Framework for Band Selection of Hy- perspectral Image,

Yaoming Cai, Xiaobo Liu, and Zhihua Cai, “BS-Nets: An End-to-End Framework for Band Selection of Hy- perspectral Image,” IEEE Transactions on Geoscience and Remote Sensing , vol. 58, no. 3, pp. 1969–1984, 2020

work page 1969
[18]

DARecNet-BS: Unsupervised Dual-Attention Reconstruction Network for Hyperspec- tral Band Selection,

Swalpa Kumar Roy, Sayantan Das, Tiecheng Song, and Bhabatosh Chanda, “DARecNet-BS: Unsupervised Dual-Attention Reconstruction Network for Hyperspec- tral Band Selection,” IEEE Geoscience and Remote Sensing Letters, vol. 18, no. 12, pp. 2152–2156, 2021

work page 2021
[19]

SRL-SOA: SELF- REPRESENTATION LEARNING WITH SPARSE 1D-OPERATIONAL AUTOENCODER FOR HY- PERSPECTRAL IMAGE BAND SELECTION,

Mete Ahishali, Serkan Kiranyaz, Iftikhar Ah- mad, and Moncef Gabbouj, “SRL-SOA: SELF- REPRESENTATION LEARNING WITH SPARSE 1D-OPERATIONAL AUTOENCODER FOR HY- PERSPECTRAL IMAGE BAND SELECTION,” in Proceedings - International Conference on Image Processing, ICIP. 2022, pp. 2296–2300, IEEE

work page 2022
[20]

A Band Selection Method with Masked Convolutional Autoencoder for Hyperspectral Image,

Yufei Liu, Xiaorun Li, Ziqiang Hua, Chaoqun Xia, and Liaoying Zhao, “A Band Selection Method with Masked Convolutional Autoencoder for Hyperspectral Image,” IEEE Geoscience and Remote Sensing Letters , vol. 19, pp. 1–5, 2022

work page 2022
[21]

Concrete autoencoders: Differentiable feature se- lection and reconstruction,

Abubakar Abid, Muhammed Fatih Balin, and James Zou, “Concrete autoencoders: Differentiable feature se- lection and reconstruction,” 36th International Confer- ence on Machine Learning, ICML 2019, vol. 2019-June, pp. 694–711, 2019

work page 2019
[22]

The concrete distribution: A continuous relaxation of discrete random variables,

Chris J. Maddison, Andriy Mnih, and Yee Whye Teh, “The concrete distribution: A continuous relaxation of discrete random variables,” 5th International Confer- ence on Learning Representations, ICLR 2017 - Confer- ence Track Proceedings, pp. 1–20, 2017

work page 2017
[23]

Categorical reparameterization with gumbel-softmax,

Eric Jang, Shixiang Gu, and Ben Poole, “Categorical reparameterization with gumbel-softmax,” 5th Interna- tional Conference on Learning Representations, ICLR 2017 - Conference Track Proceedings, pp. 1–13, 2017

work page 2017
[24]

Novel Gumbel-Softmax Trick En- abled Concrete Autoencoder with Entropy Constraints for Unsupervised Hyperspectral Band Selection,

He Sun, Jinchang Ren, Huimin Zhao, Peter Yuen, and Julius Tschannerl, “Novel Gumbel-Softmax Trick En- abled Concrete Autoencoder with Entropy Constraints for Unsupervised Hyperspectral Band Selection,” IEEE Transactions on Geoscience and Remote Sensing , vol. 60, no. June, pp. 1–13, 2022

work page 2022
[25]

Stochastic gate-based autoencoder for unsupervised hyperspectral band selection,

He Sun, Lei Zhang, Lizhi Wang, and Hua Huang, “Stochastic gate-based autoencoder for unsupervised hyperspectral band selection,” Pattern Recognition, vol. 132, pp. 108969, 2022

work page 2022
[26]

Dropout Feature Ranking for Deep Learning Models

Chun-hao Chang, Ladislav Rampasek, and Anna Gold- enberg, “Dropout Feature Ranking for Deep Learning Models,” arXiv preprint arXiv:1712.08645, 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017
[27]

A sampling,

Chris J. Maddison, Daniel Tarlow, and Tom Minka, “A sampling,” in Advances in Neural Information Process- ing Systems, 2014, vol. 4, pp. 3086–3094

work page 2014
[28]

Dropout: A simple way to prevent neural networks from overfit- ting,

Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov, “Dropout: A simple way to prevent neural networks from overfit- ting,” Journal of Machine Learning Research , vol. 15, pp. 1929–1958, 2014

work page 1929
[29]

220 Band A VIRIS Hyperspectral Image Data Set: June 12, 1992 Indian Pine Test Site 3,

Marion. F. Baumgardner, Larry. L. Biehl, and David. A. Landgrebe, “220 Band A VIRIS Hyperspectral Image Data Set: June 12, 1992 Indian Pine Test Site 3,” in Purdue University Research Repository, 2015

work page 1992
[30]

Pavia Centre and University,

Paolo Gamba, “Pavia Centre and University,” in Telecommunications and Remote Sensing Laboratory, Pavia university (Italy)

work page
[31]

Salinas,

EUDAT, “Salinas,” in EUDAT B2SHARE; http://b2share.eudat.eu, 2014

work page 2014
[32]

AmeriFlux AmeriFlux US-KS2 Kennedy Space Center (scrub oak),

Drake Bert and Hinkle Ross, “AmeriFlux AmeriFlux US-KS2 Kennedy Space Center (scrub oak),” 2016

work page 2016
[33]

PyTorch: An imperative style, high- performance deep learning library,

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas K ¨opf, Edward Yang, Zach DeVito, Martin Raison, Alykhan Tejani, Sasank Chil- amkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala, “PyTorch: An imperative style, high- pe...

work page 2019
[34]

Classification of Hyperspectral Remote Sensing,

Farid Melgani and Lorenzo Bruzzone, “Classification of Hyperspectral Remote Sensing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 42, no. August, pp. 1778–1790, 2004

work page 2004
[35]

Sparse representation based band selection for hyperspectral images,

Shuangjiang Li and Hairong Qi, “Sparse representation based band selection for hyperspectral images,” Pro- ceedings - International Conference on Image Process- ing, ICIP, pp. 2693–2696, 2011

work page 2011
[36]

Graph Convolutional Sub- space Clustering: A Robust Subspace Clustering Frame- work for Hyperspectral Image,

Yaoming Cai, Zijia Zhang, Zhihua Cai, Xiaobo Liu, Xinwei Jiang, and Qin Yan, “Graph Convolutional Sub- space Clustering: A Robust Subspace Clustering Frame- work for Hyperspectral Image,” IEEE Transactions on Geoscience and Remote Sensing , vol. 59, no. 5, pp. 4191–4202, 2021

work page 2021

[1] [1]

Dropout Concrete Autoencoder for Band Selection on HSI Scenes

INTRODUCTION Hyperspectral images (HSI) captured by hyperspectral remote sensing imaging spectrometers [1] cover a wide and continu- ous range of the electromagnetic spectrum beyond the visible wavelengths with multiple spectral bands. Due to this charac- teristic, hyperspectral images contain enormous information utilizing its various applications, such ...

work page internal anchor Pith review Pith/arXiv arXiv 2024

[2] [2]

Next, the pseudo-code of the proposed method is provided at the end of this section

PROPOSED METHOD In this section, we first present the principle of concrete dis- tribution, dropout feature ranking, and the proposed Dropout CAE in detail. Next, the pseudo-code of the proposed method is provided at the end of this section. 2.1. Concrete Distribution The concrete random variables are defined as a continuous re- laxation of discrete rando...

work page

[3] [3]

EXPERIMENTAL EV ALUA TION In this section, we first describe the datasets used in the ex- periments, and we provide implementation details which are followed by comparative evaluations. 3.1. Datasets In this work, we have evaluated the proposed network using four HSI scenes: Indian Pines [24], PaviaU [25], Salinas [26], and KSC [27]. The original HSI scen...

work page

[4] [4]

with three Q values (polynomial approximation order), SpaBS [30], EGCSR [31], ISSC [9], and BS-Net-FC [12]. Dropout CAE (T1) Dropout CAE (T2) Dropout CAE (T3) BS_Net_FC EGCSR ISSC SpaBS SRL_SOR Q1 SRL_SOR Q2 SRL_SOR Q3 0 7 15 30 45 60 75 90 105 120 135 150 165 180 195205 Spectral Band 1 2 3 4 5Value of Entropy Fig. 3: The selected subset with 20 bands and...

work page

[5] [5]

We first utilize the variational dropout strategy to exploit the importance of each frequency band for HSI scenes

CONCLUSION In this work, we propose a novel method named Dropout CAE to re-parameterize the discrete random variables for HSI band selection. We first utilize the variational dropout strategy to exploit the importance of each frequency band for HSI scenes. To bridge the gap between the discrete band information and the re-parameterization of the discrete ...

work page

[6] [6]

Ngadi and Li Liu, Hyperspectral Image Pro- cessing Techniques, Number December 2010

Michael O. Ngadi and Li Liu, Hyperspectral Image Pro- cessing Techniques, Number December 2010. Elsevier Inc., first edit edition, 2010

work page 2010

[7] [7]

Hy- perspectral image data mining for band selection in agri- cultural applications,

S. G. Bajwa, P. Bajcsy, P. Groves, and L. F. Tian, “Hy- perspectral image data mining for band selection in agri- cultural applications,” Transactions of the American So- ciety of Agricultural Engineers , vol. 47, no. 3, pp. 895– 907, 2004

work page 2004

[8] [8]

Estimation of Mineral Abundance from Hyperspectral Data Using a New Supervised Neighbor-Band Ratio Un- mixing Approach,

Kevin Siebels, Kalifa Go ¨ıta, and Micka ¨el Germain, “Estimation of Mineral Abundance from Hyperspectral Data Using a New Supervised Neighbor-Band Ratio Un- mixing Approach,” IEEE Transactions on Geoscience and Remote Sensing , vol. 58, no. 10, pp. 6754–6766, 2020

work page 2020

[9] [9]

Advances in hyperspectral image classification: Earth monitoring with statistical learning methods,

Gustavo Camps-Valls, Devis Tuia, Lorenzo Bruzzone, and Jon Atli Benediktsson, “Advances in hyperspectral image classification: Earth monitoring with statistical learning methods,” IEEE Signal Processing Magazine , vol. 31, no. 1, pp. 45–54, 2014

work page 2014

[10] [10]

Convolutional Neural Network Based on Bandwise- Independent Convolution and Hard Thresholding for Hyperspectral Band Selection,

Jie Feng, Jiantong Chen, Qigong Sun, Ronghua Shang, Xianghai Cao, Xiangrong Zhang, and Licheng Jiao, “Convolutional Neural Network Based on Bandwise- Independent Convolution and Hard Thresholding for Hyperspectral Band Selection,” IEEE Transactions on Cybernetics, vol. 51, no. 9, pp. 4414–4428, 2021

work page 2021

[11] [11]

Hyperspectral Band Selection for Mul- tispectral Image Classification with Convolutional Net- works,

Giorgio Morales, John Sheppard, Riley Logan, and Joseph Shaw, “Hyperspectral Band Selection for Mul- tispectral Image Classification with Convolutional Net- works,” Proceedings of the International Joint Confer- ence on Neural Networks , vol. 2021-July, 2021

work page 2021

[12] [12]

A joint band prioritization and band- decorrelation approach to band selection for hyperspec- tral image classification,

Chein I. Chang, “A joint band prioritization and band- decorrelation approach to band selection for hyperspec- tral image classification,” IEEE Transactions on Geo- science and Remote Sensing , vol. 37, no. 6, pp. 2631– 2641, 1999

work page 1999

[13] [13]

Band Selection for Dimension Reduction in Hyper Spectral Image Using Integrated Information- Gain and Principal Components Analysis Technique,

Kitti Koonsanit, Chuleerat Jaruskulchai, and Apisit Eiumnoh, “Band Selection for Dimension Reduction in Hyper Spectral Image Using Integrated Information- Gain and Principal Components Analysis Technique,” International Journal of Machine Learning and Com- puting, vol. 2, no. 3, pp. 248–251, 2012

work page 2012

[14] [14]

Band Selection Using Improved Sparse Subspace Clustering for Hyperspectral Imagery Classification,

Weiwei Sun, Liangpei Zhang, Bo Du, Weiyue Li, and Yenming Mark Lai, “Band Selection Using Improved Sparse Subspace Clustering for Hyperspectral Imagery Classification,” IEEE Journal of Selected Topics in Ap- plied Earth Observations and Remote Sensing , vol. 8, no. 6, pp. 2784–2797, 2015

work page 2015

[15] [15]

Optimal Clustering Framework for Hyperspectral Band Selec- tion,

Qi Wang, Fahong Zhang, and Xuelong Li, “Optimal Clustering Framework for Hyperspectral Band Selec- tion,” IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 10, pp. 5910–5922, 2018

work page 2018

[16] [16]

A Fast Neigh- borhood Grouping Method for Hyperspectral Band Se- lection,

Qi Wang, Qiang Li, and Xuelong Li, “A Fast Neigh- borhood Grouping Method for Hyperspectral Band Se- lection,” IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 6, pp. 5028–5039, 2021

work page 2021

[17] [17]

BS-Nets: An End-to-End Framework for Band Selection of Hy- perspectral Image,

Yaoming Cai, Xiaobo Liu, and Zhihua Cai, “BS-Nets: An End-to-End Framework for Band Selection of Hy- perspectral Image,” IEEE Transactions on Geoscience and Remote Sensing , vol. 58, no. 3, pp. 1969–1984, 2020

work page 1969

[18] [18]

DARecNet-BS: Unsupervised Dual-Attention Reconstruction Network for Hyperspec- tral Band Selection,

Swalpa Kumar Roy, Sayantan Das, Tiecheng Song, and Bhabatosh Chanda, “DARecNet-BS: Unsupervised Dual-Attention Reconstruction Network for Hyperspec- tral Band Selection,” IEEE Geoscience and Remote Sensing Letters, vol. 18, no. 12, pp. 2152–2156, 2021

work page 2021

[19] [19]

SRL-SOA: SELF- REPRESENTATION LEARNING WITH SPARSE 1D-OPERATIONAL AUTOENCODER FOR HY- PERSPECTRAL IMAGE BAND SELECTION,

Mete Ahishali, Serkan Kiranyaz, Iftikhar Ah- mad, and Moncef Gabbouj, “SRL-SOA: SELF- REPRESENTATION LEARNING WITH SPARSE 1D-OPERATIONAL AUTOENCODER FOR HY- PERSPECTRAL IMAGE BAND SELECTION,” in Proceedings - International Conference on Image Processing, ICIP. 2022, pp. 2296–2300, IEEE

work page 2022

[20] [20]

A Band Selection Method with Masked Convolutional Autoencoder for Hyperspectral Image,

Yufei Liu, Xiaorun Li, Ziqiang Hua, Chaoqun Xia, and Liaoying Zhao, “A Band Selection Method with Masked Convolutional Autoencoder for Hyperspectral Image,” IEEE Geoscience and Remote Sensing Letters , vol. 19, pp. 1–5, 2022

work page 2022

[21] [21]

Concrete autoencoders: Differentiable feature se- lection and reconstruction,

Abubakar Abid, Muhammed Fatih Balin, and James Zou, “Concrete autoencoders: Differentiable feature se- lection and reconstruction,” 36th International Confer- ence on Machine Learning, ICML 2019, vol. 2019-June, pp. 694–711, 2019

work page 2019

[22] [22]

The concrete distribution: A continuous relaxation of discrete random variables,

Chris J. Maddison, Andriy Mnih, and Yee Whye Teh, “The concrete distribution: A continuous relaxation of discrete random variables,” 5th International Confer- ence on Learning Representations, ICLR 2017 - Confer- ence Track Proceedings, pp. 1–20, 2017

work page 2017

[23] [23]

Categorical reparameterization with gumbel-softmax,

Eric Jang, Shixiang Gu, and Ben Poole, “Categorical reparameterization with gumbel-softmax,” 5th Interna- tional Conference on Learning Representations, ICLR 2017 - Conference Track Proceedings, pp. 1–13, 2017

work page 2017

[24] [24]

Novel Gumbel-Softmax Trick En- abled Concrete Autoencoder with Entropy Constraints for Unsupervised Hyperspectral Band Selection,

He Sun, Jinchang Ren, Huimin Zhao, Peter Yuen, and Julius Tschannerl, “Novel Gumbel-Softmax Trick En- abled Concrete Autoencoder with Entropy Constraints for Unsupervised Hyperspectral Band Selection,” IEEE Transactions on Geoscience and Remote Sensing , vol. 60, no. June, pp. 1–13, 2022

work page 2022

[25] [25]

Stochastic gate-based autoencoder for unsupervised hyperspectral band selection,

He Sun, Lei Zhang, Lizhi Wang, and Hua Huang, “Stochastic gate-based autoencoder for unsupervised hyperspectral band selection,” Pattern Recognition, vol. 132, pp. 108969, 2022

work page 2022

[26] [26]

Dropout Feature Ranking for Deep Learning Models

Chun-hao Chang, Ladislav Rampasek, and Anna Gold- enberg, “Dropout Feature Ranking for Deep Learning Models,” arXiv preprint arXiv:1712.08645, 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017

[27] [27]

A sampling,

Chris J. Maddison, Daniel Tarlow, and Tom Minka, “A sampling,” in Advances in Neural Information Process- ing Systems, 2014, vol. 4, pp. 3086–3094

work page 2014

[28] [28]

Dropout: A simple way to prevent neural networks from overfit- ting,

Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov, “Dropout: A simple way to prevent neural networks from overfit- ting,” Journal of Machine Learning Research , vol. 15, pp. 1929–1958, 2014

work page 1929

[29] [29]

220 Band A VIRIS Hyperspectral Image Data Set: June 12, 1992 Indian Pine Test Site 3,

Marion. F. Baumgardner, Larry. L. Biehl, and David. A. Landgrebe, “220 Band A VIRIS Hyperspectral Image Data Set: June 12, 1992 Indian Pine Test Site 3,” in Purdue University Research Repository, 2015

work page 1992

[30] [30]

Pavia Centre and University,

Paolo Gamba, “Pavia Centre and University,” in Telecommunications and Remote Sensing Laboratory, Pavia university (Italy)

work page

[31] [31]

Salinas,

EUDAT, “Salinas,” in EUDAT B2SHARE; http://b2share.eudat.eu, 2014

work page 2014

[32] [32]

AmeriFlux AmeriFlux US-KS2 Kennedy Space Center (scrub oak),

Drake Bert and Hinkle Ross, “AmeriFlux AmeriFlux US-KS2 Kennedy Space Center (scrub oak),” 2016

work page 2016

[33] [33]

PyTorch: An imperative style, high- performance deep learning library,

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas K ¨opf, Edward Yang, Zach DeVito, Martin Raison, Alykhan Tejani, Sasank Chil- amkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala, “PyTorch: An imperative style, high- pe...

work page 2019

[34] [34]

Classification of Hyperspectral Remote Sensing,

Farid Melgani and Lorenzo Bruzzone, “Classification of Hyperspectral Remote Sensing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 42, no. August, pp. 1778–1790, 2004

work page 2004

[35] [35]

Sparse representation based band selection for hyperspectral images,

Shuangjiang Li and Hairong Qi, “Sparse representation based band selection for hyperspectral images,” Pro- ceedings - International Conference on Image Process- ing, ICIP, pp. 2693–2696, 2011

work page 2011

[36] [36]

Graph Convolutional Sub- space Clustering: A Robust Subspace Clustering Frame- work for Hyperspectral Image,

Yaoming Cai, Zijia Zhang, Zhihua Cai, Xiaobo Liu, Xinwei Jiang, and Qin Yan, “Graph Convolutional Sub- space Clustering: A Robust Subspace Clustering Frame- work for Hyperspectral Image,” IEEE Transactions on Geoscience and Remote Sensing , vol. 59, no. 5, pp. 4191–4202, 2021

work page 2021